Any idea why they are reporting the estimated lifespan at 290°C? Testing seems to have been done at 440°C and above.
casey2•Feb 18, 2026
Coz the paper gives a function for extrapolating from these tests. This is purely testing thermal decay.
10,000 years sounds like a good benchmark and isn't as obviously ridiculous as saying a million years at 260°C
idiotsecant•Feb 21, 2026
It's common to perform longevity testing at higher temperatures to simulate longer lifetimes, in account of nobody has decades of time to actually perform a 1x time test.
HPsquared•Feb 21, 2026
I wonder if "damp" modes of decay could still damage them though, which isn't captured in this style of testing. Like some wet chemical or biological process.
jurgenburgen•Feb 21, 2026
Mechanical decay would also damage them. I think it’s assumed that the media will be stored in a place protected from humidity, chemicals and hammers.
HPsquared•Feb 21, 2026
Yes I suppose a strong casing can protect against all that, but not against temperature so that's the one thing they still need to test for.
That's ~7-18 days per 120mm x 120mm medium (4.8TB).
Glass prices stable for now. Also, the authors make no statement about horizontal vs. vertical storage.
NitpickLawyer•Feb 21, 2026
Thanks for digging this up. Every "scientists create new storage medium" is always a disappointment when you get to see the write speeds. This seems decent? At least in "raw" numbers there's nothing obviously making this useless. Let's hope they have a path to quick commercialisation and make it available. If there's any DC adoption will be the real test, I think.
stackghost•Feb 21, 2026
>This seems decent?
Definitely. If it actually achieves those speeds it's perfectly reasonable for long-term/cold storage.
Someone•Feb 21, 2026
Depends somewhat on the read speed, too. Extreme example: if that is one bit per year, it doesn’t matter that you can write stuff on it.
dyauspitr•Feb 21, 2026
I imagine if you can use lasers to etch at that speed, you can use them to read at similar speeds as well.
po1nt•Feb 21, 2026
First CDs would take hour and a half to write with a laser. Once engineers take over the tech, it will might get faster.
wumms•Feb 21, 2026
If they get the read speed up to a couple of GBit/s (~100x current max write speed), 4.8TB might be a good fit for 32k movies.
thegrim33•Feb 21, 2026
Write speed is probably the least important metric for people that are considering something like this. After everything with storage and longevity is taken care of, improving write speeds is a nice to have, but not the important part.
Am4TIfIsER0ppos•Feb 21, 2026
> No read speed given
Write only medium!
npodbielski•Feb 21, 2026
At least it is safe for 10k years! And from everybody ever basically.
jmclnx•Feb 18, 2026
The big question, is it patented to the point were no one can buy the burners and media ?
Will it run on Linux ?
misswaterfairy•Feb 21, 2026
They're definitely pursuing patents...
> The authors of the paper have filed several patents relating to the subject matter contained in this paper in the name of Microsoft Corporation.
It's whether Microsoft will be fair and flexible licensing their patents to third-parties.
Otherwise I'd suggest that if they keep it all to themselves and charge like a wounded bull, uptake would be quite limited.
At least until the original patents expires, which might be the better strategic move for third-parties in light of a hostile Microsoft given how long this archival format is expected to last.
HPsquared•Feb 21, 2026
20 year patent lifespan seems like nothing to the overall lifespan of this invention though.
nandomrumber•Feb 21, 2026
Or, next year I can buy ten for five dollars off AliExpress.
vasco•Feb 21, 2026
Yeah but then 1000 years from now nobody will have the right USB cable to read it.
I think we should stick to proven solutions for millennia-robust information storage and paint it on walls inside pyramids.
stackghost•Feb 21, 2026
If Nanni could have engraved his shitpost about Ea-nasir's copper into multiple glass tablets, easy to distribute, that would last for 10000 years, he probably would have.
userbinator•Feb 21, 2026
You could "bootstrap" all the information required to produce the hardware to read this, by starting with human-readable instructions for the next step.
dguest•Feb 21, 2026
What language will humans be reading in 10,000 years?
Aardwolf•Feb 21, 2026
You don't necessarily need the same hardware to read it, just like you can read a vinyl record optically without a needle
dyauspitr•Feb 21, 2026
Depends on what you etch on there. If it’s binary representation of actual alphabets then sure. If it’s a video file then without the software to decipher and manipulate the data, it would be pretty indecipherable. How to read an mp4 is not part of the data itself.
simicd•Feb 21, 2026
At 4.8TB one could add a header section with the full code, instructions how to compile it etc. That would certainly help to reproduce it, assuming civilizations in 10k years still can decypher todays language.
ksec•Feb 21, 2026
>4.84TB in a single slab of glass, (the slabs are 12 cm x 12 cm and 0.2 cm thick).
So a rough estimate, at the size of UMD, used in Playstation Portable, slightly smaller than the size of Mini Disc, it could store 1TB.
I assume we could do double layer in the future for 2TB.
For comparison that is roughly 1000x times the capacity of UMD. I would love to have this. Burn a few of these as backup and call it a day.
Karliss•Feb 21, 2026
It is already way beyond double layer. The 4.8TB is achieved using 301 layers.
ksec•Feb 21, 2026
There goes my hope of non-cloud backup. I was thinking 1TB doesn't quite make it. Or at least I need a dozens of these.
rarisma•Feb 21, 2026
I swear this happens at least once a year.
Wheres my futuristic storage guys?
winrid•Feb 21, 2026
in your hands :)
idiotsecant•Feb 21, 2026
Glass is one of the more stable things we can make. This seems pretty good! I don't have an application that requires ten thousand years of storage but I'm sure someone out there does!
tbrownaw•Feb 21, 2026
> I don't have an application that requires ten thousand years of storage but I'm sure someone out there does!
A) record (a representative cross-section of) "everything" and leave multiple copies where future archeologists might find it. To avoid things like how present-day archaeologists apparently have holes in the kinds of things they can find, due to different social classes not leaving equally-robust trails.
B) this is "at least as long as I could possibly care about" storage. If I need to retain say financial records for seven years, and then later the government retroactively increases that to 20 years, there's no need to re-archive it all onto new media.
TacticalCoder•Feb 21, 2026
> B) this is "at least as long as I could possibly care about" storage. If I need to retain say financial records for seven years, and then later the government retroactively increases that to 20 years, there's no need to re-archive it all onto new media.
In many countries this "maximum (6 or) 7 years" for financial records is only if the local IRS decides that you're not potentially committing fraud. If they decide you've potentially committed fraud at any time in the past, there's no limit as to how far they can go. Even in the US stuff like (some of the) funds stolen by the Enron scam have been successfully clawed back more than two decades after the fact.
At least that's the case in several EU countries: there's literally no limit if the country's IRS equivalent decides you're potentially committing fraud (or if you did in the past).
Which is insane and totally arbitrary but that's how it is.
In addition to that under a great many KYC/AML excuses, there are banks out there that shall have zero issue asking you to justify the "source of funds" and at times I've had to provide info dating from way more than seven years in the past. I've heard --and I'm not shitting you-- from someone proving he bought for about 5 K EUR of something that went up more than 100x (think Bitcoin or some exceptionally successful stock), that his bank answered something like: "OK, but now that you've proven you actually made 100x, prove us the source of the 5 K EUR in 2013!".
That's what happens to a society when you give too much power to petty people.
There are literally collaborationists out there that are going to fill SARs (Suspicious Activity Reports) when someone can prove he turned 5 K into 500 K not on the 500 K (which are impossible to dispute) but on the 5 K that were used in the first place. That's how jealous and incompetent some people are in this world.
Things became so bad that I now have a Git versioned repo (and backups everywhere) where I keep track of, among other, every single wire transfer above 10 K EUR. I've got stuff dating back to 2001 when I bought my first apartment etc.
Don't underestimate how pathetic and bitter some of the people you'll have to deal with (be it from your local IRS or a bank) are going to be.
lencastre•Feb 21, 2026
given a long enough period, glass is a fluid, i.e. viscosity
You can get a 6mmx6mmx1.2mm pure industrial diamond sheet for about $1000 [1]. That should be able to hold around 300 GB with this method and would last practically forever.
This is mostly a myth based on some medieval glass panels that had structurally wider bases. This material is going to take until the heat death of the universe to deform 1mm at room temperature. I'm sure it'll be fine for way longer than it will take for the data to fail in a different way.
gigel82•Feb 21, 2026
I have read a variation of this headline once every 2 years since the early 2000s, yet never seen it turn into something real (that a consumer / enterprise can buy).
> Microsoft began to build on their work in 2017. Although Kazansky’s approach maximizes durability and the density of data, in the latest work, Microsoft has gone for practicality. They explore a method that enables data to be written faster and decoded more reliably than did Project Silica’s previous iterations, says Black, and it uses cheaper borosilicate glass, rather than harder-to-make fused silica.
Something I've always wondered.. could things like tree rings, or birdsong and other mating rituals (the species can survive only by replicating the sequences) be used for "data storage" by some hypothetical intelligences that think on scales of thousands of years?
alfiedotwtf•Feb 21, 2026
I don’t trust my Kodak Gold CDs that advertised 40 years, let alone humans to not self annihilate within the next 400 years.
In fact, look what we’re doing right now with all our past’s relics!
theAdminWave•Feb 21, 2026
LOL I've done holographic data storage in borosilicate glass using fs laser pulses for my masters thesis in physics more than a decade ago and guess what, this is not going anywhere. The claims are all wildly exaggerated also. Lots of buzzwords micro nano plasma explosions but the truth is hidden in the details: needs specialist hardware... Yeah like a 50.000 USD femto second laser setup that needs an entire basement and you wearing ski googles at all times to not get blind type of specialist hardware. Guess we're all gonna put that in our living rooms, won't we?
And the storage density is limited by all kinds of effects that I won't even get into it but you can roughly assume its at best half or even less of that and then it starts becoming much less impressive.
Yes you can microwave a slab of glass or go diving with it and it will still be intact but unless we make machines that read and store data much more easily, like significantly absurdly more easily, this is the biggest pipe dream of them all.
Cool tech though :)
baxtr•Feb 21, 2026
Agreed. On the other hand: didn’t any cool tech start as a overpriced, oversized version of its later breakthrough product?
stanac•Feb 21, 2026
It doesn't have to be consumer hardware to be economically viable. I can imagine something like this replacing or complementing tape storage at data centers. We already have hard drives filled with gas for dust-proofing. For archival storage it does not have to be fast (in terms of latency) it just needs to be reliable with high data density.
Hard drives where the size of a car decades ago, we could now have archival storage of the same physical size that can hold petabytes (just guessing, didn't do the actual math).
fragmede•Feb 21, 2026
The question is how much are readers. If I have to take my data to my local strip mall to write my datacube, but this data cube can be read with a much cheaper reader that I can reasonably believe will be available in 40 years, I could see that as being viable.
zkmon•Feb 21, 2026
Data loses its usefulness and relevance with time. Unless it's updatable, just archiving is not going to be useful. Just like how we don't find some 2000 years old writings any useful now, except for museum storage.
b0rtb0rt•Feb 21, 2026
pretty sure there are a few 2000 year old texts that people are still reading today
bubblewand•Feb 21, 2026
Pretty sure the oldest stuff I've read was around 4,500 years old (in translation, but still). Volume I of Lichtheim's Ancient Egyptian Literature.
There are a couple incomplete tales in that 3-volume work and I really wish they'd had a more stable storage medium, because now I'm stuck with accidental millennia-old cliffhangers.
thaumasiotes•Feb 21, 2026
We've had data storage that lasts for millennia for several thousand years already. The invention of millennia-long storage more or less coincided with the invention of writing.
There isn't really a benefit. Our durably-stored several-thousand-year-old records suffer from various problems:
- They're hard to understand.
- They tend not to be relevant to much.
- Most of them have gotten lost. They're not gone, but it would be extremely expensive to find them.
Interestingly, these are the same problems that occur with stored data of much more recent vintage. But they get worse and worse over time, and the fact that the storage medium itself doesn't degrade does nothing to help. It tends to make those usability problems worse by giving people a false sense of security that the data is still there, until the cost of recovering it becomes too great and for practical purposes it isn't there anymore.
If something matters, it will be stored on ephemeral media and recopied over time onto more ephemeral media.
20 Comments
10,000 years sounds like a good benchmark and isn't as obviously ridiculous as saying a million years at 260°C
Definitely. If it actually achieves those speeds it's perfectly reasonable for long-term/cold storage.
Write only medium!
Will it run on Linux ?
> The authors of the paper have filed several patents relating to the subject matter contained in this paper in the name of Microsoft Corporation.
Page 12 of the paper: https://www.nature.com/articles/s41586-025-10042-w.pdf
It's whether Microsoft will be fair and flexible licensing their patents to third-parties.
Otherwise I'd suggest that if they keep it all to themselves and charge like a wounded bull, uptake would be quite limited.
At least until the original patents expires, which might be the better strategic move for third-parties in light of a hostile Microsoft given how long this archival format is expected to last.
I think we should stick to proven solutions for millennia-robust information storage and paint it on walls inside pyramids.
So a rough estimate, at the size of UMD, used in Playstation Portable, slightly smaller than the size of Mini Disc, it could store 1TB.
I assume we could do double layer in the future for 2TB.
For comparison that is roughly 1000x times the capacity of UMD. I would love to have this. Burn a few of these as backup and call it a day.
Wheres my futuristic storage guys?
A) record (a representative cross-section of) "everything" and leave multiple copies where future archeologists might find it. To avoid things like how present-day archaeologists apparently have holes in the kinds of things they can find, due to different social classes not leaving equally-robust trails.
B) this is "at least as long as I could possibly care about" storage. If I need to retain say financial records for seven years, and then later the government retroactively increases that to 20 years, there's no need to re-archive it all onto new media.
In many countries this "maximum (6 or) 7 years" for financial records is only if the local IRS decides that you're not potentially committing fraud. If they decide you've potentially committed fraud at any time in the past, there's no limit as to how far they can go. Even in the US stuff like (some of the) funds stolen by the Enron scam have been successfully clawed back more than two decades after the fact.
At least that's the case in several EU countries: there's literally no limit if the country's IRS equivalent decides you're potentially committing fraud (or if you did in the past).
Which is insane and totally arbitrary but that's how it is.
In addition to that under a great many KYC/AML excuses, there are banks out there that shall have zero issue asking you to justify the "source of funds" and at times I've had to provide info dating from way more than seven years in the past. I've heard --and I'm not shitting you-- from someone proving he bought for about 5 K EUR of something that went up more than 100x (think Bitcoin or some exceptionally successful stock), that his bank answered something like: "OK, but now that you've proven you actually made 100x, prove us the source of the 5 K EUR in 2013!".
That's what happens to a society when you give too much power to petty people.
There are literally collaborationists out there that are going to fill SARs (Suspicious Activity Reports) when someone can prove he turned 5 K into 500 K not on the 500 K (which are impossible to dispute) but on the 5 K that were used in the first place. That's how jealous and incompetent some people are in this world.
Things became so bad that I now have a Git versioned repo (and backups everywhere) where I keep track of, among other, every single wire transfer above 10 K EUR. I've got stuff dating back to 2001 when I bought my first apartment etc.
Don't underestimate how pathetic and bitter some of the people you'll have to deal with (be it from your local IRS or a bank) are going to be.
https://www.scientificamerican.com/article/fact-fiction-glas...
[1] https://e6cvd.com/us/material/single-crystalline.html?utm_so...
Now, can someone please help me get some data of this Iomega ZIPdrive disc?
I thought I was experiencing some Mandela affect, had to Bing it. This is from 2022 https://www.microsoft.com/en-us/research/video/project-silic...
> Microsoft began to build on their work in 2017. Although Kazansky’s approach maximizes durability and the density of data, in the latest work, Microsoft has gone for practicality. They explore a method that enables data to be written faster and decoded more reliably than did Project Silica’s previous iterations, says Black, and it uses cheaper borosilicate glass, rather than harder-to-make fused silica.
Following your link, I found a prototype of the media storage system (2023) with just 2828 views: https://www.youtube.com/watch?v=xnK-uB4OsgU
Something I've always wondered.. could things like tree rings, or birdsong and other mating rituals (the species can survive only by replicating the sequences) be used for "data storage" by some hypothetical intelligences that think on scales of thousands of years?
In fact, look what we’re doing right now with all our past’s relics!
And the storage density is limited by all kinds of effects that I won't even get into it but you can roughly assume its at best half or even less of that and then it starts becoming much less impressive.
Yes you can microwave a slab of glass or go diving with it and it will still be intact but unless we make machines that read and store data much more easily, like significantly absurdly more easily, this is the biggest pipe dream of them all.
Cool tech though :)
Hard drives where the size of a car decades ago, we could now have archival storage of the same physical size that can hold petabytes (just guessing, didn't do the actual math).
There are a couple incomplete tales in that 3-volume work and I really wish they'd had a more stable storage medium, because now I'm stuck with accidental millennia-old cliffhangers.
There isn't really a benefit. Our durably-stored several-thousand-year-old records suffer from various problems:
- They're hard to understand.
- They tend not to be relevant to much.
- Most of them have gotten lost. They're not gone, but it would be extremely expensive to find them.
Interestingly, these are the same problems that occur with stored data of much more recent vintage. But they get worse and worse over time, and the fact that the storage medium itself doesn't degrade does nothing to help. It tends to make those usability problems worse by giving people a false sense of security that the data is still there, until the cost of recovering it becomes too great and for practical purposes it isn't there anymore.
If something matters, it will be stored on ephemeral media and recopied over time onto more ephemeral media.